Chat GPT AI

wormie · Feb 4, 2023

Captain Suave said:
He's misunderstanding the role of randomness and conflating concepts from the training process with the final model's output process, thinking that you could create a large distribution of outputs and somehow aggregate/filter them for "correctness", this getting a better output at additional computational cost. But that's not how this works.

I think my recommendation of ML for dummies was an excellent idea. Followed by DL for dummies!

pharmakos · Feb 4, 2023

wormie said:
"The incorrect output, while it happens, is rare"

The incorrect output is always there and its never rare. Without millions of incorrect outputs, the model would not have any sort of accuracy. Unless you mean incorrect output from chatGPT itself? Those arent part of the training data set.

We must be using different definitions of the word "rare" because the thing can pass the MBA.

I was not at any point talking about changing the training process. I was completely talking about a way to curate the final output with the current model. I have no idea what I said that made you think I was talking about changing the training model but at no point was it what I was talking about.

wormie said:
"the classic model for machine learning being one algorithm that is good at error checking things, the other that keeps trying to adjust its own model until it can fool the error checker"

This is not the classical model of machine learning. This is a model of a encoder-decoder architecture for a deep learning models.

Classical was the wrong term perhaps but it is definitely the model that the majority of AI learning algorithms are following these days.

wormie said:
"seeds"

Wut?

"aggregate average"

Wut?

You are far more ignorant than you let on if this is confusing to you. Behind the scenes it's math all the way down. The reason there's a different output each time you put in the same input is because behind the scenes it's using a different random seed. Are you really this ignorant while pretending to be this smart? Because holy shit wormie. I always knew you were arrogant and stupid but this is new levels for you.

I'm not even gonna bother reading the rest of that comment. You're a moron.

pharmakos · Feb 4, 2023

Actually nah I can't help myself, I'll keep going

wormie said:
anomalous output should be removed"

Outputs are not removed during training. Removing output from chatGPT responses is meaningless as it is outside of training.

I never said shit was getting removed from the training data by the algorithm. I was talking about the final output that the user sees. You're grasping at straws to make me look dumb but you don't even understand what the fuck I'm saying. Largely because you are far more ignorant than you realize.

wormie said:
"And obviously the concept works better for large language models than for image generation AI."

Why is it obvious? You are talking about adjusting training inputs which are numerical vectors in image recognization as they are in language recognition. Yeah, in one they are binary and in the other they are gray scale rgb values but still, what is being adjusted and what exactly is obvious about anything you are saying here?

Captain Suave understood it I'm sure. It is obvious. On the spectrum from deterministic to probablistic, noise diffusion is way more probabilistic and large language models are way more deterministic. Which is because you WANT image generation to offer a plethora of results that are then curated, as one of the primary reasons, but also just going back to the way a computer encodes images versus the way it encodes text. Think of the difference in size between a txt file and a high res jpg if that works better for your peanut brain.

wormie said:
"My proposal would involve generating something like 10-50 outputs for each input"

What output are you talking about? What input? Computing power isnt limited, just add more GPUs. Time might become a constraint but if you are training a model for a month, you can afford to train for 2 months.

If computing power wasn't limited, then users wouldn't be waiting in queues to log on. Are you seriously this fucking stupid? Yeah sure you can scale it up infinitely if you have infinite dollars. No shit Sherlock. We are talking about the real world tho.

pharmakos · Feb 4, 2023

Captain Suave said:
He's misunderstanding the role of randomness and conflating concepts from the training process with the final model's output process, thinking that you could create a large distribution of outputs and somehow aggregate/filter them for "correctness", thus getting a better output at additional computational cost. Basically some kind of ML perpetual accuracy machine.

I wasn't saying anything about the training process. I was completely just talking about ways to adjust the final output. What I was saying involved zero changes to the training data.

Hateyou · Feb 4, 2023

Sanrith Descartes · Feb 4, 2023

How did this thread go from GPT outputs making fun of Ossoi to this?

This is why we can't have nice things.

Hateyou · Feb 4, 2023

Sanrith Descartes said:
How did this thread go from GPT outputs making fun of Ossoi to this?

This is why we can't have nice things.

Captain Suave · Feb 4, 2023

pharmakos said:
I wasn't saying anything about the training process. I was completely just talking about ways to adjust the final output. What I was saying involved zero changes to the training data.

Yes, I get that.

Look, I'm not personally a ML expert but I've looked over the shoulder of one for 15 years and built a handful of smaller-scale models myself. I understand what you are describing and I promise you that it's just not how this stuff works. Wormie appears to know what he's talking about, and the reason you think he's an idiot is because he's actually doing you the favor of connecting your half-baked idea to real contact points in the development process.

Anyway, I'm out on this debate. Feel free to carry on wondering why no one is already doing this; I can only refer you to previous replies. (No, it's not compute.)

pharmakos · Feb 4, 2023

Captain Suave said:
Yes, I get that.

Look, I'm not personally a ML expert but I've looked over the shoulder of one for 15 years and built a handful of smaller-scale models myself. I understand what you are describing and I promise you that it's just not how this stuff works. Wormie appears to know what he's talking about, and the reason you think he's an idiot is because he's actually doing you the favor of connecting your half-baked idea to real contact points in the development process.

Anyway, I'm out on this debate. Feel free to carry on wondering why no one is already doing this; I can only refer you to previous replies. (No, it's not compute.)

He doesn't even know what a random seed is.

pharmakos · Feb 4, 2023

Literally just lazily pasting the first three links after typing "random seed in ai" into google here:

Why use Random Seed in Machine Learning? - Analytics Yogi

Data, Data Science, Machine Learning, Deep Learning, Analytics, Python, R, Tutorials, Tests, Interviews, News, AI

vitalflux.com

https://link.medium.com/fZW8yefe9wb

Properly Setting the Random Seed in ML Experiments. Not as Simple as You Might Imagine

Join Comet at Booth 406 in the ODSC East Expo Hall. We will also be speaking at ODSC: – April 30, 9 am — A Deeper Stack for Deep Learning: Adding Visualizations + Data Abstractions to your Workflow (Douglas Blank, Head of Research) [Training Session] – May 2, 2:15 pm —...

opendatascience.com

Daidraco · Feb 4, 2023

Captain Suave said:
Yes, I get that.

Look, I'm not personally a ML expert but I've looked over the shoulder of one for 15 years and built a handful of smaller-scale models myself. I understand what you are describing and I promise you that it's just not how this stuff works. Wormie appears to know what he's talking about, and the reason you think he's an idiot is because he's actually doing you the favor of connecting your half-baked idea to real contact points in the development process.

Anyway, I'm out on this debate. Feel free to carry on wondering why no one is already doing this; I can only refer you to previous replies. (No, it's not compute.)

Dont leave!!! You can make it all better by replying with Chat GPT!

wormie · Feb 4, 2023

pharmakos said:
Literally just lazily pasting the first three links after typing "random seed in ai" into google here:

Why use Random Seed in Machine Learning? - Analytics Yogi

Data, Data Science, Machine Learning, Deep Learning, Analytics, Python, R, Tutorials, Tests, Interviews, News, AI

vitalflux.com

https://link.medium.com/fZW8yefe9wb

Properly Setting the Random Seed in ML Experiments. Not as Simple as You Might Imagine

Join Comet at Booth 406 in the ODSC East Expo Hall. We will also be speaking at ODSC: – April 30, 9 am — A Deeper Stack for Deep Learning: Adding Visualizations + Data Abstractions to your Workflow (Douglas Blank, Head of Research) [Training Session] – May 2, 2:15 pm —...

opendatascience.com

Those links are talking about setting the seed for the pseudo random number generators in order to gain exact output of models instead of random one. This is done when building anything that relies on randomness in order to verify correctness. These links have nothing to do with what you were talking about.. Shut the fuck up nimrod,

The other post I am not going to address except to say that the reason you think Vanessa was a good poster is because he too argued about subjects he knew nothing about. You are a non fag version of Vanessa and even that is questionable.

pharmakos · Feb 4, 2023

wormie said:
Those links are talking about setting the seed for the pseudo random number generators in order to gain exact output of models instead of random one. This is done when building anything that relies on randomness in order to verify correctness. These links have nothing to do with what you were talking about.. Shut the fuck up nimrod,

Right, they're about setting exact random seeds instead of letting the computer decide. If you don't put in your own random seed, the computer uses an algorithm to generate a pseudorandom one (often based on things like server time or other variables -- think back to people who learned to game /random in EverQuest if that helps you understand).

pharmakos · Feb 4, 2023

Just admit you had no idea that machine learning involved hidden variable random seeds and we can move on instead of shitting up yet another thread where people assume that just because of my reputation that I must be the one that's wrong. You are wrong. Machine learning output involves a random seed for every prompt. You're just a moron.

wormie · Feb 4, 2023

pharmakos said:
Just admit you had no idea that machine learning involved hidden variable random seeds and we can move on instead of shitting up yet another thread where people assume that just because of my reputation that I must be the one that's wrong. You are wrong. Machine learning involves a random seed for every prompt. You're just a moron.

“Shouldn't checking multiple random seeds against each other and then producing/assuming the result with the fewest errors based on the averages there end up outputting far more accurate results?”

No shit machine learning uses a seed to generate random values. Anything that generates such values uses a seed. A seed is not set manually any time one actually trains machine learning models. Setting a random seed to a value results in exactly the same output every time anything random is done. If the random seed you were talking about is the setting for the pseudo number generator then you are even a bigger retard than we all think you are. You are literally saying that we should remove randomness from models when training and compare exact outputs and accept these supposedly probabilistic results as such. If this is thr case then you are a massive fucking retard. But no one is this stupid, not even you, so you are just a retard that read a blog posts and thinks now he knows something and is actually fucking arguing about it when it’s obvious to anyone that you are talking ou of your ass. Go away.

Daidraco · Feb 4, 2023

Hateyou said:

pharmakos · Feb 4, 2023

wormie said:
No shit machine learning uses a seed to generate random values.

Your earlier posts contained things like "lol what is a seed?" You clearly are backpedaling. Not going to bother reading the rest of that comment. I think I might just finally put you on ignore.

wormie · Feb 4, 2023

pharmakos said:
Your earlier posts contained things like "lol what is a seed?" You clearly are backpedaling. Not going to bother reading the rest of that comment. I think I might just finally put you on ignore.

Thats because "seed" made no sense in the context of what you were saying you fucking retard. Shut the fuck up.

pharmakos · Feb 4, 2023

wormie said:
Thats because "seed" made no sense in the context of what you were saying you fucking retard. Shut the fuck up.

It would have made sense if you knew what the fuck you were talking about. Each individual response to a prompt has a random seed attached. Trying multiple seeds and filtering out the portions of each seed that don't match the average output will lead to more accurate results. Make sense now? It just would take multiplicatively more processing power and they already can't keep up with user demand.

TomServo · Feb 4, 2023

wormie said:
“Shouldn't checking multiple random seeds against each other and then producing/assuming the result with the fewest errors based on the averages there end up outputting far more accurate results?”

No shit machine learning uses a seed to generate random values. Anything that generates such values uses a seed. A seed is not set manually any time one actually trains machine learning models. Setting a random seed to a value results in exactly the same output every time anything random is done. If the random seed you were talking about is the setting for the pseudo number generator then you are even a bigger retard than we all think you are. You are literally saying that we should remove randomness from models when training and compare exact outputs and accept these supposedly probabilistic results as such. If this is thr case then you are a massive fucking retard. But no one is this stupid, not even you, so you are just a retard that read a blog posts and thinks now he knows something and is actually fuck youing arguing about it when it’s obvious to anyone that you are talking ou of your ass. Go away.

Now I hope he starts talking out his ass about IVs and encryption

Chat GPT AI

Asshat wormie

2023 Asshat Award Winner

soʞɐɯɹɐɥd

soʞɐɯɹɐɥd

soʞɐɯɹɐɥd

Not Great, Not Terrible

Veteran of a thousand threadban wars

Not Great, Not Terrible

Caesar si viveret, ad remum dareris.

soʞɐɯɹɐɥd

soʞɐɯɹɐɥd

Golden Baronet of the Realm

Asshat wormie

2023 Asshat Award Winner

soʞɐɯɹɐɥd

soʞɐɯɹɐɥd

Asshat wormie

2023 Asshat Award Winner

Golden Baronet of the Realm

soʞɐɯɹɐɥd

Asshat wormie

2023 Asshat Award Winner

soʞɐɯɹɐɥd